DOCSLIB.ORG

Better Image Annotation with Sub-Image Decomposition

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Better Image Annotation with Sub-Image Decomposition Block Annotation: Better Image Annotation with Sub-Image Decomposition Hubert Lin Paul Upchurch Kavita Bala Cornell University Cornell University Cornell University Abstract are required for principled conclusions to be drawn. In the past, polygon annotation tools have enabled partially dense Image datasets with high-quality pixel-level annotations annotations (in which small semantic regions are densely are valuable for semantic segmentation: labelling every annotated) to be crowdsourced at scale with public crowd pixel in an image ensures that rare classes and small ob- workers. These tools paved the way for the cost-effective cre- jects are annotated. However, full-image annotations are ation of large-scale partially dense datasets such as [8, 37]. expensive, with experts spending up to 90 minutes per image. Despite the success of these annotation tools, fully dense We propose block sub-image annotation as a replacement datasets have relied extensively on expensive expert annota- for full-image annotation. Despite the attention cost of fre- tors [60, 14, 41, 64, 42] and private crowdworkers [11]. quent task switching, we find that block annotations can We propose annotation of small blocks of pixels as a be crowdsourced at higher quality compared to full-image stand-in replacement for full-image annotation (figure1). annotation with equal monetary cost using existing annota- We find that these annotations can be effectively gathered tion tools developed for full-image annotation. Surprisingly, by crowdworkers, and that annotation of a sparse number of we find that 50% pixels annotated with blocks allows se- blocks per image can train a high performance segmentation mantic segmentation to achieve equivalent performance to network. We further show these sparsely annotated images 100% pixels annotated. Furthermore, as little as 12% of can be extended automatically to full-image annotations. pixels annotated allows performance as high as 98% of the We show block annotation has: performance with dense annotation. In weakly-supervised • Wide applicability. (Section3) Block annotations can be settings, block annotation outperforms existing methods by effectively crowdsourced at higher quality compared to full 3-4% (absolute) given equivalent annotation time. To re- annotation. It is easy to implement and works with existing cover the necessary global structure for applications such as advances in image annotation. characterizing spatial context and affordance relationships, we propose an effective method to inpaint block-annotated • Cost-efficient Design. (Section3) Block annotation re- images with high-quality labels without additional human flects a cost-efficient design paradigm (while current re- effort. As such, fewer annotations can also be used for these search focuses on reducing annotation time). This is remi- applications compared to full-image annotation. niscent of gamification and citizen science where enjoyable tasks lead to low-cost high-engagement work. • Complex Region Annotation. (Section3) Block annota- 1. Introduction tion shifts focus from categorical regions to spatial regions. When annotating categorical regions, workers segment sim- Recent large-scale computer vision datasets place a heavy ple objects before complex objects. With spatial regions, emphasis on high-quality fully dense annotations (in which informative complex regions are forced to be annotated. over 90% of the pixels are labelled) for hundreds of thou- • Weakly-Supervised Performance. (Section4) Block an- sands of images. Dense annotations are valuable for both notation is competitive in weakly-supervised settings, outper- semantic segmentation and applications beyond segmenta- forming existing methods by 3-4% (absolute) given equiva- tion such as characterizing spatial context and affordance lent annotation time. relationships [11, 23]. The long-tail distribution of classes means it is difficult to gather annotations for rare classes, es- • Scalable Performance. (Section4) Full-supervision per- pecially if these classes are difficult to segment. Annotating formance is achieved by annotating 50% of blocks per image. every pixel in an image ensures that pixels corresponding to Thus, blocks can be annotated until desired performance is rare classes or small objects are labelled. Dense annotations achieved, in contrast to methods such as scribbles. also capture pixels that form the boundary between classes. • Scalable Structure. (Section5) Block-annotated images For applications such as understanding spatial context be- can be effectively inpainted with high quality labels without tween classes or affordance relationships, dense annotations additional human effort. 1 Figure 1: (a) Sub-image block annotations are more effective to gather than full-image annotations (b) Training on sparse block annotations enables semantic segmentation performance equivalent to full-image annotations (c) Block labels can be inpainted with high-quality labels. 2. Related Work human-annotated blocks can be extended automatically into dense annotations (sec.5), and we discuss how other human- In this section we review recent works on pixel-level anno- machine methods can be used with blocks (sec. 3.6). tation in three areas: human annotation, and human-machine annotation, and dense segmentation with weak supervision. Weakly-Supervised Dense Segmentation. There are al- Human Annotation. Manual labeling of every pixel is ternatives to training with high-quality densely annotated impractical for large-scale datasets. A successful method images which substitute quantity for label quality and/or rich- is to have crowdsource workers segment polygonal regions ness. Previous works have used low-quality pixel-level an- to click on boundaries. Employing crowdsource workers notations [65], bounding boxes [45, 28, 49], point-clicks [7], offers its own set of challenges with quality control and task scribbles [7, 36], image-level class labels [45, 53, 3], image- design [56, 8, 55]. Although large-scale public crowdsourc- level text descriptions [24] and unlabeled related web ing can be successful [37] recent benchmark datasets have videos [24] to train semantic segmentation networks. Com- resorted to in-house expert annotators [14, 43]. Annotation bining weak annotations with small amounts of high-quality time can be reduced through improvements such as autopan, dense annotation is another strategy for reducing cost [9, 26]. zoom [8] and shared polygon boundaries [60]. Polygon seg- [52] proposes a two-stage approach where image-level class mentation can be augmented by painted labels on superpixel labels are automatically converted into pixel-level masks groups [11] and Bezier curves [60]. Pixel-level labels for which are used to train a semantic segmentation network. images can also be obtained by (1) constructing a 3D scene We find a small number of sub-image block annotations is a from an image collection, (2) grouping and labeling 3D competitive form of weak supervision (sec. 4.3). shapes and (3) propagating shape labels to image pixels [39]. In our work, we investigate sub-image polygon annotation, which can be further combined with other methods (sec.3.) 3. Block Annotation Human-Machine Annotation. Complex boundaries are Sub-image block annotation is composed of three stages: time-consuming to trace manually. In these cases the cost of (1) Given an image I, select a small spatial region I0; (2) pixel-level annotation can be reduced by automating a por- Annotate I0 with pixel-level labels; (3) Repeat (with different tion of the task. Matting and object selection [50, 33, 34, 6, I0) until I is sufficiently annotated. In this paper, we explore 58, 57, 10, 30, 59] generate tight boundaries from loosely an- the case where I0 is rectangular, and focus on the use of notated boundaries or few inside/outside clicks and scribbles. existing pixel-level annotation tools. [44, 38] introduced a predictive method which automatically Can block annotations be gathered as effectively as full- infers a foreground mask from 4 boundary clicks, and was image annotations with existing tools? In section 3.1, we extended to full-image segmentation in [2]. The number show our annotation interface. In section 3.2, we explore the of boundary clicks was further reduced to as few as one quality of block annotation versus full-image annotation. In by [1]. Predictive methods require an additional human section 3.3, we examine block annotation for a real-world verification step since the machine mutates the original hu- dataset. In section 3.4, we discuss the cost of block annota- man annotation. The additional step can be avoided with an tion and show worker feedback. In section 3.5, we discuss online method. However, online methods (e.g., [1, 30, 2]) how blocks for annotation can be selected in practice. Fi- have higher requirements since the algorithm must be trans- nally, in section 3.6 we discuss the compatibility of block lated into the web browser setting and the worker’s machine annotation with existing annotation methods. must be powerful enough to run the algorithm1. Alterna- tively, automatic proposals can be generated for humans to manipulate: [5] generates segments, [4] generates a set of 3.1. Annotation Interface matting layers, [61] generates superpixel labels, and [47] Our block annotation interface is given in figure2 and generates boundary fragments. In our work, we show that implemented with existing tools [8]. For full image annota- 1Offloading online methods onto a cloud service offers a different
Load more

Recommended publications
  • What's the Point: Semantic Segmentation with Point Supervision
    What's the Point: Semantic Segmentation with Point Supervision Amy Bearman1, Olga Russakovsky2, Vittorio Ferrari3, and Li Fei-Fei1 1 Stanford University fabearman,[email protected] 2 Carnegie Mellon University [email protected] 3 University of Edinburgh [email protected] Abstract. The semantic image segmentation task presents a trade-off between test time accuracy and training-time annotation cost. Detailed per-pixel annotations enable training accurate models but are very time- consuming to obtain; image-level class labels are an order of magnitude cheaper but result in less accurate models. We take a natural step from image-level annotation towards stronger supervision: we ask annotators to point to an object if one exists. We incorporate this point supervision along with a novel objectness potential in the training loss function of a CNN model. Experimental results on the PASCAL VOC 2012 benchmark reveal that the combined effect of point-level supervision and object- ness potential yields an improvement of 12:9% mIOU over image-level supervision. Further, we demonstrate that models trained with point- level supervision are more accurate than models trained with image-level, squiggle-level or full supervision given a fixed annotation budget. Keywords: semantic segmentation, weak supervision, data annotation 1 Introduction At the forefront of visual recognition is the question of how to effectively teach computers new concepts. Algorithms trained from carefully annotated data enjoy better performance than their weakly supervised counterparts (e.g., [1] vs. [2], [3] vs. [4], [5] vs. [6]), yet obtaining such data is very time-consuming [5, 7]. It is particularly difficult to collect training data for semantic segmentation, i.e., the task of assigning a class label to every pixel in the image.
    [Show full text]
  • Feature Mining: a Novel Training Strategy for Convolutional Neural
    Feature Mining: A Novel Training Strategy for Convolutional Neural Network Tianshu Xie1 Xuan Cheng1 Xiaomin Wang1 [email protected] [email protected] [email protected] Minghui Liu1 Jiali Deng1 Ming Liu1* [email protected] [email protected] [email protected] Abstract In this paper, we propose a novel training strategy for convo- lutional neural network(CNN) named Feature Mining, that aims to strengthen the network’s learning of the local feature. Through experiments, we find that semantic contained in different parts of the feature is different, while the network will inevitably lose the local information during feedforward propagation. In order to enhance the learning of local feature, Feature Mining divides the complete feature into two com- plementary parts and reuse these divided feature to make the network learn more local information, we call the two steps as feature segmentation and feature reusing. Feature Mining is a parameter-free method and has plug-and-play nature, and can be applied to any CNN models. Extensive experiments demonstrate the wide applicability, versatility, and compatibility of our method. 1 Introduction Figure 1. Illustration of feature segmentation used in our method. In each iteration, we use a random binary mask to Convolution neural network (CNN) has made significant divide the feature into two parts. progress on various computer vision tasks, 4.6, image classi- fication10 [ , 17, 23, 25], object detection [7, 9, 22], and seg- mentation [3, 19]. However, the large scale and tremendous parameters of CNN may incur overfitting and reduce gener- training strategy for CNN for strengthening the network’s alizations, that bring challenges to the network training.
    [Show full text]
  • Arxiv:2004.07999V4 [Cs.CV] 23 Jul 2021
    REVISE: A Tool for Measuring and Mitigating Bias in Visual Datasets Angelina Wang · Alexander Liu · Ryan Zhang · Anat Kleiman · Leslie Kim · Dora Zhao · Iroha Shirai · Arvind Narayanan · Olga Russakovsky Abstract Machine learning models are known to per- petuate and even amplify the biases present in the data. However, these data biases frequently do not become apparent until after the models are deployed. Our work tackles this issue and enables the preemptive analysis of large-scale datasets. REVISE (REvealing VIsual bi- aSEs) is a tool that assists in the investigation of a visual dataset, surfacing potential biases along three Fig. 1: Our tool takes in as input a visual dataset and its dimensions: (1) object-based, (2) person-based, and (3) annotations, and outputs metrics, seeking to produce geography-based. Object-based biases relate to the size, insights and possible actions. context, or diversity of the depicted objects. Person- based metrics focus on analyzing the portrayal of peo- ple within the dataset. Geography-based analyses con- sider the representation of different geographic loca- the visual world, representing a particular distribution tions. These three dimensions are deeply intertwined of visual data. Since then, researchers have noted the in how they interact to bias a dataset, and REVISE under-representation of object classes (Buda et al., 2017; sheds light on this; the responsibility then lies with Liu et al., 2009; Oksuz et al., 2019; Ouyang et al., 2016; the user to consider the cultural and historical con- Salakhutdinov et al., 2011; J. Yang et al., 2014), ob- text, and to determine which of the revealed biases ject contexts (Choi et al., 2012; Rosenfeld et al., 2018), may be problematic.
    [Show full text]
  • AI for All by Mr. S. Arjun
    AI for All Mr. S. Arjun 2nd Year, B.Tech CSE (Hons.), Lovely Professional University, Punjab [email protected] What is AI Many problems in the world that once seemed impossible for a computer to tackle without human intervention are solved today with Artificial Intelligence. We are witnessing the second major wave of AI, disrupting a plethora of unrelated fields such as health, ethics, politics, and economy. These intelligent systems prove that machines too can learn from experience, adapt, and make meaningful decisions. While the first wave was driven by rule-based systems where experts in performing a task handcrafted a set of rules for machines to follow, thus emulating intelligence, the second wave is driven by huge amounts of data, coupled with algorithms that enable machines to recognize patterns and learn from experience. Impact Though at a very early stage, AI has already made a profound impact on our lives. However, the nature of the impact it has made on different establishments is as unique as itself. On Enterprises Enterprises are seen to make the best out of the second wave of AI, primarily owing to the abundance of data they already collect every second. Deep Learning - a subset of Machine Learning, that allows recognizing and learning from complex patterns in data - feeds on huge magnitudes of data in the order of millions of samples. Large enterprises have the benefit of being capable of collecting such data in-house for their own systems, products, and services. Today, numerous such organizations that operate at scale are starting to use AI to automate workflows, streamline their operations, and optimize production.
    [Show full text]
  • Crowdsourcing in Computer Vision
    Foundations and Trends R in Computer Graphics and Vision Vol. 10, No. 3 (2014) 177–243 c 2016 A. Kovashka, O. Russakovsky, L. Fei-Fei and K. Grauman DOI: 10.1561/0600000073 Crowdsourcing in Computer Vision Adriana Kovashka Olga Russakovsky University of Pittsburgh Carnegie Mellon University [email protected] [email protected] Li Fei-Fei Kristen Grauman Stanford University University of Texas at Austin [email protected] [email protected] Contents 1 Introduction 178 2 What annotations to collect 181 2.1 Visual building blocks . 182 2.2 Actions and interactions . 191 2.3 Visual story-telling . 198 2.4 Annotating data at different levels . 204 3 How to collect annotations 205 3.1 Interfaces for crowdsourcing and task managers . 205 3.2 Labeling task design . 207 3.3 Evaluating and ensuring quality . 211 4 Which data to annotate 215 4.1 Active learning . 215 4.2 Interactive annotation . 221 5 Conclusions 226 References 228 ii Abstract Computer vision systems require large amounts of manually annotated data to properly learn challenging visual concepts. Crowdsourcing plat- forms offer an inexpensive method to capture human knowledge and un- derstanding, for a vast number of visual perception tasks. In this survey, we describe the types of annotations computer vision researchers have collected using crowdsourcing, and how they have ensured that this data is of high quality while annotation effort is minimized. We begin by discussing data collection on both classic (e.g., object recognition) and recent (e.g., visual story-telling) vision tasks. We then summarize key design decisions for creating effective data collection interfaces and workflows, and present strategies for intelligently selecting the most important data instances to annotate.
    [Show full text]
  • Cornernet-Lite- Efficient Keypoint Based Object Detection
    H. LAW, Y. TANG, O. RUSSAKOVSKY, J. DENG: CORNERNET-LITE 1 CornerNet-Lite: Efficient Keypoint-Based Object Detection Hei Law Department of Computer Science [email protected] Princeton University Yun Tang Princeton, NJ USA [email protected] Olga Russakovsky [email protected] Jia Deng [email protected] Abstract Keypoint-based methods are a relatively new paradigm in object detection, eliminat- ing the need for anchor boxes and offering a simplified detection framework. Keypoint- based CornerNet achieves state of the art accuracy among single-stage detectors. How- ever, this accuracy comes at high processing cost. In this work, we tackle the problem of efficient keypoint-based object detection and introduce CornerNet-Lite. CornerNet-Lite is a combination of two efficient variants of CornerNet: CornerNet-Saccade, which uses an attention mechanism to eliminate the need for exhaustively processing all pixels of the image, and CornerNet-Squeeze, which introduces a new compact backbone architecture. Together these two variants address the two critical use cases in efficient object detection: improving efficiency without sacrificing accuracy, and improving accuracy at real-time efficiency. CornerNet-Saccade is suitable for offline processing, improving the efficiency arXiv:1904.08900v2 [cs.CV] 16 Sep 2020 of CornerNet by 6.0x and the AP by 1.0% on COCO. CornerNet-Squeeze is suitable for real-time detection, improving both the efficiency and accuracy of the popular real-time detector YOLOv3 (34.4% AP at 30ms for CornerNet-Squeeze compared to 33.0% AP at 39ms for YOLOv3 on COCO). Together these contributions for the first time reveal the potential of keypoint-based detection to be useful for applications requiring processing efficiency.
    [Show full text]
  • Deep Learning with Spherical Data Min Sun National Tsing Hua University Vision Science Lab Self-Intro
    Deep Learning with Spherical Data Min Sun National Tsing Hua University Vision Science Lab http://aliensunmin.github.io/ Self-Intro 遠見專訪 https://www.gvm.com.tw/article.html?id=43046 Large-scale video analysis Video Title Generation Video QA Visual Forecasting ECCV 2016 AAAI 2017 ICCV 2017 ACCV 2016 Best Paper Award CVPR 2017 Car Accident Anticipation Risk Assessment Handcam Object Recognition via Handcam Grasp Assistant for BVI ECCV 2016 Under Submission 2018 ICCV 2017 Intention Anticipation 360 (Spherical) Vision Deep Pilot Saliency Prediction CVPR 2017 CVPR 2018 AAAI 2018 360 Grounding RoBotic Vision Learn from Demonstration Indoor Place Recognition & Navigation ICRA 2018 Robust Control Health Care Outline •Introduction to Deep Learning •Spherical data •Deep Learning for Spherical data Deep Learning (DL) • Data • GPU Computing • Talents Data: • 開始於 2007 @ Princeton • 初登場於 2009 @ CVPR • 照⽚停⽌搜集於 2010 Ø總共類別： 21841 Jia Deng Fei-Fei Li Ø總共圖⽚： 1千 4百萬 • ILSVR Challenge 從 2010到現今 Info from http://www.image-net.org/ 1K Image Classification Label = f(Image) Deep Learning 深度學習 Figure from Olga Russakovsky ECCV'14 workshop GPU: NVIDIA CUDA Tesla P100 With Over 20 TFLOPS Of FP16 Read more: http://wccftech.com/nvidia-pascal-gpu-gtc-2016/#ixzz456KT75Jf FaceBook Machine Learning Server Big Sur: 8 NVIDIA M40 https://techcrunch.com/gallery/a-look-inside-facebooks-data-center Talents: DNNresearch acquired By Google Geoffrey Hinton (right: Professor) Alex Krizhevsky (middle; PhD student), and Ilya Sutskever (left; Postdoc) Deep Learning: Convolutional Neural
    [Show full text]
  • WHEN SMART MACHINES ARE BIASED Olga Russakovsky Is Working to Change That
    SEXUAL-MISCONDUCT TRANSFER STUDENTS PRINCETON’S RECOMMENDATIONS ADMITTED DIGITAL ARCHIVES PRINCETON ALUMNI WEEKLY WHEN SMART MACHINES ARE BIASED Olga Russakovsky is working to change that JUNE 6, 2018 PAW.PRINCETON.EDU 00paw0606-coverFINALrev1.indd 1 5/22/18 10:23 AM That moment you finally get to go through the gates and see the next horizon. Communications of UNFORGETTABLE Office PRINCETON Your support makes it possible. This year’s Annual Giving campaign ends on June 30, 2018. To contribute by credit card, or for more information please call the gift line at 800-258-5421 (outside the US, 609-258-3373), or visit www.princeton.edu/ag. June 6, 2018 Volume 118, Number 14 An editorially independent magazine by alumni for alumni since 1900 PRESIDENT’S PAGE 2 INBOX 3 ON THE CAMPUS 5 Sexual-misconduct recommendations Transfer students Maya Lin visit Class Close-up: Video games deal with climate change A Day With ... dining hall coordinator Manuel Gomez Castaño ’20 Frank Stella ’58 exhibition at PUAM SPORTS: International rowers Championship for women’s crew The Big Three LIFE OF THE MIND 17 Author and professor Yiyun Li Rick Barton on peacemaking PRINCETONIANS 29 Mallika Ahluwalia ’05 Q&A: Yolanda Pierce ’94, first woman to lead Howard University divinity school Tiger Caucus in Congress CLASS NOTES 32 Maya Lin, page 7 Wikipedia Bias and Artificial Intelligence 20 Born Digital 24 ’20; MEMORIALS 49 Olga Russakovsky draws on personal With the decline of paper records, the staff at CLASSIFIEDS 54 experience and technical expertise to help Mudd Library is ensuring that digital records make artificial intelligence smarter.
    [Show full text]
  • Masters Thesis: Development of a Deep Learning Model for 3D Human Pose Estimation in Monocular Videos
    Development of a Deep Learning Model for 3D Human Pose Estimation in Monocular Videos Agnė Grinciūnaitė Master’s Degree Thesis VILNIUS GEDIMINAS TECHNICAL UNIVERSITY Faculty of Fundamental Sciences Department of Graphical Systems Agnė Grinciūnaitė Development of a Deep Learning Model for 3D Human Pose Estimation in Monocular Videos Master’s degree Thesis Information Technologies study programme, state code 621E14004 Multimedia Information Systems specialization Informatics Engineering study field Vilnius, 2016 The work in this thesis was supported by Vicar Vision. Their cooperation is hereby grate- fully acknowledged. Copyright © Department of Graphical Systems All rights reserved. VILNIUS GEDIMINAS TECHNICAL UNIVERSITY Faculty of Fundamental Sciences Department of Graphical Systems APPROVED BY Head of Department (Signature) (Name, Surname) (Date) Agnė Grinciūnaitė Development of a Deep Learning Model for 3D Human Pose Estimation in Monocular Videos Master’s degree Thesis Information Technologies study programme, state code 621E14004 Multimedia Information Systems specialization Informatics Engineering study field Supervisor (Title, Name, Surname) (Signature) (Date) Consultant (Title, Name, Surname) (Signature) (Date) Consultant (Title, Name, Surname) (Signature) (Date) Vilnius, 2016 Abstract There exists a visual system which can easily recognize and track human body position, movements and actions without any additional sensing. This system has the processor called brain and it is competent after being trained for some months. With a little bit more training it is also able to apply acquired skills for more complicated tasks such as understanding inter-personal attitudes, intentions and emotional states of the observed moving person. This system is called a human being and is so far the most inspirational piece of art for today’s artificial intelligence creators.
    [Show full text]
  • Computer Vision and Its Implications Based on the Paper Imagenet Classification with Deep Convolutional Neural Networks by Alex Krizhevsky, Ilya Sutskever, Geoffrey E
    Universität Leipzig Fakultät für Mathematik und Informatik Institut für Informatik Problem-seminar Deep-Learning Summary about Computer Vision and its implications based on the paper ImageNet Classification with Deep Convolutional Neural Networks by Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton Leipzig, WS 2017/18 vorgelegt von Christian Stur Betreuer: Ziad Sehili Contents 1.0. Introduction ..................................................................................................................... 2 1.1. Present significance ........................................................................................................ 2 1.2. Motivation of Computer Vision ...................................................................................... 2 2.0. Computer Vision ............................................................................................................. 3 2.1. Neural Networks ............................................................................................................. 3 2.2. Large Scale Visual Recognition Challenge (ILSVRC) .................................................. 4 2.2.1. Goal .......................................................................................................................... 4 2.2.2. Challenge Progression ............................................................................................. 4 2.2.3. The Data ................................................................................................................... 6 2.2.4. Issues
    [Show full text]
  • A Study of Face Obfuscation in Imagenet
    A Study of Face Obfuscation in ImageNet Kaiyu Yang 1 Jacqueline Yau 2 Li Fei-Fei 2 Jia Deng 1 Olga Russakovsky 1 Abstract devices taking photos (Butler et al., 2015; Dai et al., 2015). Face obfuscation (blurring, mosaicing, etc.) has Learning from the visual data has led to computer vision been shown to be effective for privacy protection; applications that promote the common good, e.g., better nevertheless, object recognition research typically traffic management (Malhi et al., 2011) and law enforce- assumes access to complete, unobfuscated images. ment (Sajjad et al., 2020). However, it also raises privacy In this paper, we explore the effects of face concerns, as images may capture sensitive information such obfuscation on the popular ImageNet challenge as faces, addresses, and credit cards (Orekondy et al., 2018). visual recognition benchmark. Most categories in Preventing unauthorized access to sensitive information in the ImageNet challenge are not people categories; private datasets has been extensively studied (Fredrikson however, many incidental people appear in the et al., 2015; Shokri et al., 2017). However, are publicly images, and their privacy is a concern. We first available datasets free of privacy concerns? Taking the annotate faces in the dataset. Then we demon- popular ImageNet dataset (Deng et al., 2009) as an example, strate that face blurring—a typical obfuscation there are only 3 people categories1 in the 1000 categories technique—has minimal impact on the accuracy of ImageNet Large Scale Visual Recognition Challenge of recognition models. Concretely, we benchmark (ILSVRC) (Russakovsky et al., 2015); nevertheless, the multiple deep neural networks on face-blurred dataset exposes many people co-occurring with other objects images and observe that the overall recognition in images (Prabhu & Birhane, 2021), e.g., people sitting accuracy drops only slightly (≤ 0:68%).
    [Show full text]
  • On Imagenet Roulette and Machine Vision
    Humans Categorise Humans: on ImageNet Roulette and Machine Vision Olga Goriunova Published in Donaufestival: Redefining Arts Catalogue, April 2020. In September 2019, artist Trevor Paglen and researcher Kate Crawford were getting through the final week of their ImageNet Roulette being publicly available on the web. ImageNet Roulette is part of their exhibition “Training Humans,” which ran between September 2019 and February 2020 at the Foundazione Prada. With this exhibition, collaborators Crawford and Paglen queried the collections of images and the processes used to train a wide range of machine learning algorithms to recognise and label images (known as “machine vision”).1 Crawford and Paglen want to draw attention to how computers see and categorize the world: seeing and recognizing is not neutral for humans (one need only think about “seeing” race or gender), and the same goes for machines. Machine vision, by now a large and successful part of AI, has only taken off in the last decade. For “machines” to recognize images, they need to be “trained” to do so. The first major step for this to happen is to have a freely available and pre-labelled very large image-data-set. The first and largest training dataset that Crawford and Paglen focused on and which they derived the title of their project from is ImageNet, which is only ten years old. ImageNet Roulette is a website or an app that allows one to take a selfie and run it through ImageNet. It uses an “open-source Caffe deep-learning framework … trained on the images and labels in the ‘person’ categories.”2 One is “seen” and “recognized” or labelled.
    [Show full text]